An Artificial Neural Network-based Approach to Predict the R-curve of Composite DCB Multidirectional Laminates

An artificial neural network-based approach is used to determine the R-curve of multidirectional laminates. The main idea of the approach is to extract the hidden information of the R-curve from the load-displacement curve of mode I delamination test while without measuring the delamination growth length. In order to obtain the training data set, R-curve is randomly generated, and then the corresponding load-displacement curve is obtained through finite element simulation. A bilinear cohesive constitutive law taking into account the R-curve is used, which has been shown to reproduce well the experiments. After training the neural network with simulated data, the load-displacement data are then taken as the input of the artificial neural network, and the description parameters of R-curve are the output. The predicted R-curves from the trained neural network are consistent with experimental results of composite laminates with different material systems and interfaces. The bridging stress and load-displacement response also agree well with experimental results. All these demonstrate applicability of the proposed approach.

Automated summary

An artificial neural network-based approach is used to determine the R-curve of multidirectional laminates. The approach extracts hidden information of the R-curve from the load-displacement curve of mode I delamination test without measuring the delamination growth length. After training the neural network with simulated data, the load-displacement data are taken as input and the parameters of R-curve are the output. The predicted R-curves from the trained neural network are consistent with experimental results, demonstrating the applicability of this approach.